China already has two exascale supercomputers-Zimo News


The Exascale era is the next holy grail of supercomputers, and it is also the guarantee of national scientific and technological progress to obtain this computing power.

If Western countries continue to make progress in the field of supercomputers, China is also developing rapidly, hoping to no longer rely on Western technology.

This topic is subtle, because the United States prevents Chinese companies from accessing the most sensitive technologies, but the Middle Eastern empire, which is silent about progress, may soon emerge from the shadows.

website Next platform Said that at least two Chinese supercomputers passed the exaflop standard in 2021, and these results will be announced soon.This will be the system Ocean stone Located in the center of Wuxi, Sunway, Taihuzhiguang supercomputer evolution ranked fourth the 500 strongest, And will reach a peak of 1.3 exaflop and a sustained performance of 1.05 exaflop at a power of 35 MW.

Another system called Tianhe Three, It will also reach a peak of 1.3 exaflop, which can maintain exascale power. It is based on Feiteng’s Feiteng processor, which uses the ARM architecture, which is not necessarily the most advanced, but is supported by a matrix accelerator.

No extraordinary buildings, but a steady rise in power

Either way, processing power will be generated through hardware optimization rather than a breakthrough new architecture, said Next platform.

The publication is also concerned that China may not publish the results it has obtained on the LINPACK benchmark, which is a benchmark for the top 500 supercomputers, which may devalue the ranking, which is still crucial to determining the development status of this market segment.

Currently, the number one is Japanese supercomputer Fugaku, with 416 petaflops (so, from a perspective, it is 0.416 exaflop).

Since the advent of supercomputers, the United States has not been excluded from the exascale level. border The personnel from Oak Ridge National Laboratory must ensure that the peak power for sustained performance is 1.5 exaflops and 1.3 exaflops.

For its part, Intel just announced the future system aurora It is expected that by 2022, the Xeon chip and Ponte Vecchio accelerator (Intel Xe-HPC architecture) in the Argonne National Laboratory will be able to climb to 2 exaflops.

Source link


Please enter your comment!
Please enter your name here